1. LLM still cannot reason in the long context.   2. The role of false positive rate. 3. Why does LLM performance degrade with context length? Really? 4. How to deal with the problem? 5. Why is the problem easily overlooked? Kuhnian structure of AI. Read more